pxfm/exponents/
exp10f.rs

1/*
2 * // Copyright (c) Radzivon Bartoshyk 6/2025. All rights reserved.
3 * //
4 * // Redistribution and use in source and binary forms, with or without modification,
5 * // are permitted provided that the following conditions are met:
6 * //
7 * // 1.  Redistributions of source code must retain the above copyright notice, this
8 * // list of conditions and the following disclaimer.
9 * //
10 * // 2.  Redistributions in binary form must reproduce the above copyright notice,
11 * // this list of conditions and the following disclaimer in the documentation
12 * // and/or other materials provided with the distribution.
13 * //
14 * // 3.  Neither the name of the copyright holder nor the names of its
15 * // contributors may be used to endorse or promote products derived from
16 * // this software without specific prior written permission.
17 * //
18 * // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
19 * // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
21 * // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
22 * // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24 * // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
25 * // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
26 * // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29use crate::exponents::expf::{ExpfBackend, GenericExpfBackend};
30
31pub(crate) struct ExpBReduc {
32    pub(crate) hi: f64,
33    pub(crate) lo: f64,
34}
35
36const MID_BITS: u32 = 5;
37const MID_MASK: usize = (1 << MID_BITS) - 1;
38const LOG2_B: f64 = f64::from_bits(0x400a934f0979a371) * (1 << MID_BITS) as f64;
39const M_LOGB_2_HI: f64 = f64::from_bits(0xbfd34413509f8000) / (1 << MID_BITS) as f64;
40const M_LOGB_2_LO: f64 = f64::from_bits(0x3d380433b83b532a) / (1 << MID_BITS) as f64;
41const EXP_2_MID: [u64; 32] = [
42    0x3ff0000000000000,
43    0x3ff059b0d3158574,
44    0x3ff0b5586cf9890f,
45    0x3ff11301d0125b51,
46    0x3ff172b83c7d517b,
47    0x3ff1d4873168b9aa,
48    0x3ff2387a6e756238,
49    0x3ff29e9df51fdee1,
50    0x3ff306fe0a31b715,
51    0x3ff371a7373aa9cb,
52    0x3ff3dea64c123422,
53    0x3ff44e086061892d,
54    0x3ff4bfdad5362a27,
55    0x3ff5342b569d4f82,
56    0x3ff5ab07dd485429,
57    0x3ff6247eb03a5585,
58    0x3ff6a09e667f3bcd,
59    0x3ff71f75e8ec5f74,
60    0x3ff7a11473eb0187,
61    0x3ff82589994cce13,
62    0x3ff8ace5422aa0db,
63    0x3ff93737b0cdc5e5,
64    0x3ff9c49182a3f090,
65    0x3ffa5503b23e255d,
66    0x3ffae89f995ad3ad,
67    0x3ffb7f76f2fb5e47,
68    0x3ffc199bdd85529c,
69    0x3ffcb720dcef9069,
70    0x3ffd5818dcfba487,
71    0x3ffdfc97337b9b5f,
72    0x3ffea4afa2a490da,
73    0x3fff50765b6e4540,
74];
75
76// Approximating 10^dx with degree-5 minimax polynomial generated by Sollya:
77// > Q = fpminimax((10^x - 1)/x, 4, [|D...|], [-log10(2)/2^6, log10(2)/2^6]);
78// Then:
79//   10^dx ~ P(dx) = 1 + COEFFS[0] * dx + ... + COEFFS[4] * dx^5.
80pub(crate) const EXP10F_COEFFS: [u64; 5] = [
81    0x40026bb1bbb55515,
82    0x40053524c73bd3ea,
83    0x4000470591dff149,
84    0x3ff2bd7c0a9fbc4d,
85    0x3fe1429e74a98f43,
86];
87
88/// Range reduction function equivalent to exp_b_range_reduc
89#[inline(always)]
90pub(crate) fn exp_b_range_reduc<B: ExpfBackend>(x: f32, backend: &B) -> ExpBReduc {
91    let xd = x as f64;
92
93    // kd = round(log2(b) * x)
94    let kd = backend.round(LOG2_B * xd);
95    let k = unsafe { kd.to_int_unchecked::<i32>() }; // it's already not indeterminate.
96
97    // hi = floor(kd / 2^MID_BITS)
98    let exp_hi = (k.wrapping_shr(MID_BITS) as u64).wrapping_shl(52); // 52 = fraction bits in f64
99
100    // mh = 2^hi * 2^mid
101    let mid_index = (k as usize) & MID_MASK;
102    let mh_bits = EXP_2_MID[mid_index].wrapping_add(exp_hi);
103    let mh = f64::from_bits(mh_bits);
104
105    // dx = x - (hi + mid) * log(2)
106    let z0 = backend.fma(kd, M_LOGB_2_HI, xd);
107    let dx = backend.fma(kd, M_LOGB_2_LO, z0);
108
109    ExpBReduc { lo: dx, hi: mh }
110}
111
112#[inline(always)]
113fn exp10f_gen<B: ExpfBackend>(x: f32, backend: B) -> f32 {
114    let x_u = x.to_bits();
115    let x_abs = x_u & 0x7fff_ffff;
116
117    // When |x| >= log10(2^128), or x is nan
118    if x_abs >= 0x421a209bu32 {
119        if x_u.wrapping_shl(1) >= 0xffu32 << 24 {
120            // |x| == inf, |x| == NaN
121            if x.is_sign_negative() && x.is_infinite() {
122                return 0.0;
123            } else if x.is_infinite() {
124                return f32::INFINITY;
125            }
126            return x + f32::NAN; // x == NaN
127        }
128        // When x < log10(2^-150) or nan
129        if x_u > 0xc2349e35u32 {
130            // exp(-Inf) = 0
131            if x.is_infinite() {
132                return 0.0;
133            }
134            // exp(nan) = nan
135            if x.is_nan() {
136                return x;
137            }
138            return 0.0;
139        }
140        // x >= log10(2^128) or nan
141        if x > 0. && (x_u >= 0x421a209bu32) {
142            // x is +inf or nan
143            return x + f32::INFINITY;
144        }
145    }
146
147    if x_abs <= 0x3d000000u32 {
148        // |x| < 1/32
149        if x_abs <= 0x3b9a209bu32 {
150            if x_u == 0xb25e5bd9u32 {
151                // x = -1.2943e-08
152                return 1.;
153            }
154            // |x| < 2^-25
155            // 10^x ~ 1 + log(10) * x
156            if x_abs <= 0x32800000u32 {
157                return backend.fmaf(x, f32::from_bits(0x40135da2), 1.0);
158            }
159        }
160
161        let xd = x as f64;
162
163        // Special polynomial for small x.
164        // Generated by Sollya:
165        // d = [-1/32, 1/32];
166        // f_exp10f = (10^y - 1)/y;
167        // Q = fpminimax(f_exp10f, 6, [|D...|], d, relative, floating);
168
169        // See ./notes/exp10f_small.sollya
170        let p = backend.polyeval7(
171            xd,
172            f64::from_bits(0x40026bb1bbb55516),
173            f64::from_bits(0x40053524c73cfbf6),
174            f64::from_bits(0x4000470591de0b07),
175            f64::from_bits(0x3ff2bd760599f3a5),
176            f64::from_bits(0x3fe142a001511a6f),
177            f64::from_bits(0x3fca7feffa781d53),
178            f64::from_bits(0x3fb16e53492c0f0e),
179        );
180        return backend.fma(p, xd, 1.) as f32;
181    }
182
183    // Range reduction: 10^x = 2^(mid + hi) * 10^lo
184    //   rr = (2^(mid + hi), lo)
185    let rr = exp_b_range_reduc(x, &backend);
186
187    // The low part is approximated by a degree-5 minimax polynomial.
188    // 10^lo ~ 1 + COEFFS[0] * lo + ... + COEFFS[4] * lo^5
189    let lo2 = rr.lo * rr.lo;
190    // c0 = 1 + COEFFS[0] * lo
191    let c0 = backend.fma(rr.lo, f64::from_bits(EXP10F_COEFFS[0]), 1.0);
192    // c1 = COEFFS[1] + COEFFS[2] * lo
193    let c1 = backend.fma(
194        rr.lo,
195        f64::from_bits(EXP10F_COEFFS[2]),
196        f64::from_bits(EXP10F_COEFFS[1]),
197    );
198    // c2 = COEFFS[3] + COEFFS[4] * lo
199    let c2 = backend.fma(
200        rr.lo,
201        f64::from_bits(EXP10F_COEFFS[4]),
202        f64::from_bits(EXP10F_COEFFS[3]),
203    );
204    // p = c1 + c2 * lo^2
205    //   = COEFFS[1] + COEFFS[2] * lo + COEFFS[3] * lo^2 + COEFFS[4] * lo^3
206    let p = backend.fma(lo2, c2, c1);
207    // 10^lo ~ c0 + p * lo^2
208    // 10^x = 2^(mid + hi) * 10^lo
209    //      ~ mh * (c0 + p * lo^2)
210    //      = (mh * c0) + p * (mh * lo^2)
211    backend.fma(p, lo2 * rr.hi, c0 * rr.hi) as f32
212}
213
214#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
215#[target_feature(enable = "avx", enable = "fma")]
216unsafe fn exp10f_fma_impl(x: f32) -> f32 {
217    use crate::exponents::expf::FmaBackend;
218    exp10f_gen(x, FmaBackend {})
219}
220
221/// Computes exp10
222///
223/// Max found ULP 0.49999508
224#[inline]
225pub fn f_exp10f(x: f32) -> f32 {
226    #[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
227    {
228        exp10f_gen(x, GenericExpfBackend {})
229    }
230    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
231    {
232        use std::sync::OnceLock;
233        static EXECUTOR: OnceLock<unsafe fn(f32) -> f32> = OnceLock::new();
234        let q = EXECUTOR.get_or_init(|| {
235            if std::arch::is_x86_feature_detected!("avx")
236                && std::arch::is_x86_feature_detected!("fma")
237            {
238                exp10f_fma_impl
239            } else {
240                fn def_exp10f(x: f32) -> f32 {
241                    exp10f_gen(x, GenericExpfBackend {})
242                }
243                def_exp10f
244            }
245        });
246        unsafe { q(x) }
247    }
248}
249
250#[cfg(test)]
251mod tests {
252    use super::*;
253
254    #[test]
255    fn test_exp10f() {
256        assert!(f_exp10f(f32::from_bits(0x7fc0_0000)).is_nan());
257        assert_eq!(f_exp10f(-1. / 64.), 0.9646616);
258        assert_eq!(f_exp10f(1. / 64.), 1.0366329);
259        assert_eq!(f_exp10f(1.), 10.0);
260        assert_eq!(f_exp10f(2.), 100.0);
261        assert_eq!(f_exp10f(3.), 1000.0);
262        assert_eq!(f_exp10f(f32::INFINITY), f32::INFINITY);
263        assert_eq!(f_exp10f(f32::NEG_INFINITY), 0.);
264        assert!(f_exp10f(f32::NAN).is_nan());
265    }
266}
pxfm/exponents/exp10f.rs

pxfm/exponents/
exp10f.rs
